Selected Projects

This project is to investigate the techniques that use textures to enhance the exhibition of surface details in 3D computer generated images. It is generally recognised that one of the major differences between real world images and computer synthesized images lies in the exhibition of high-frequency visible surface details. This project will be an effort to bridge this major gap. The principal objective of this project is to develop a key technique called Amplifiable Bi-directional Texture Function (ABTF). ABTF is designed to generate textures on 3D surface models. Compared with the conventional texture mapping/synthesis techniques, ABTF will have the following strengths: 1) It will not compromise on texture quality in close-up views; 2) It will accurately display significant visual effects of fine surface details, including self-shadowing and occlusion, inter-reflection and silhouettes, under different viewing/lighting settings. 3) It will take into consideration the influence from the variations of target surface curvatures for the correct synthesis and display of textures. The ABTF synthesis will be based on a hybrid image modelling and rendering approach. Given multiple views of a texture sample, it will recover the underlying geometries of the texture and use them as surface details for the target surface. For high visual fidelity purpose, the colours of the target surface will be obtained from the images of the multiple views through a quick search scheme for achieving high quality and fast performance. The project has a wide range of potential applications. In fact, the range of use of 3D high fidelity images in different businesses and fields is surprisingly broad, which suggests a wide range of possible commercial application. Many 3D-related businesses have significant presence in the UK, and are playing an active role in the global market, e.g. computer games, computer animation, broadcast television, mobile communication, web design, etc., all of which are facing demands for improved image quality to take full advantage of continual advances in display technology.
Researchers: F Dong and D Chen (Brunel).
Support: EPSRC
Duration: 2005-2008

Reconstructing Background of DNA Microarray Images

DNA microarray technology has enabled biologists to study all the genes within an entire organism to obtain a global view of gene interaction and regulation. However, the technology is still early in its development, and errors may be introduced at each of the main stages of the microarray process: spotting, hybridisation, and scanning. Consequently the microarray image data collected often contain errors and noise, which will then be propagated down through all later stages of processing and analysis. Although there is recently much research on how to detect and eliminate these variations and errors, the progress has been slow. The proposed project explores novel methods for processing microarray image data by reconstructing background noise of the microarray chip. It brings together expertise from the disparate fields of image processing, data mining and molecular biology to make an interdisciplinary attempt in advancing the state of art in this important area. It is particularly timely since there is an urgent need to have image analysis software that can save both time and labour as well as provide high-quality image data.
Researchers: X Liu, Y Li, Z Wang and K Fraser (Brunel), and P Kellam (UCL).
Support: EPSRC
Duration: 2006-2008

Video Annotation Using Human Activities and Visual Context

This proposal aims to develop a system for video annotation, i.e. assigning meaningful semantic labels to video units. As opposed to many previous studies where ad-hoc concepts such as 'grasses', 'sky' and 'explosion' are adopted, we will address the problem from a different perspective by combining human activity and visual context. On the one hand, humans are usually the subjects of video semantics, e.g. the doer of an action. Their presence, activities and interactions are often the key factors to video contents. On the other hand, context, the physical or informative environment or situation where human activities are undertaken, can greatly clarify ambiguity and reduce complexity in video content understanding. Based on our previous work on human face processing and semantic video analysis, we will develop new algorithms and methods for appearance based non-rigid object (e.g. face) tracking, incremental and robust person-specific facial model updating, and unsupervised automatic contextual analysis. The system will detect and track human faces and provide probabilistic descriptions of each individual human face as well as their trajectories in a video. It will also formulate person-specific facial appearance models online by incrementally and robustly updating a generic facial model. Meanwhile, the system will perform unsupervised visual context analysis from low-level features on each video segments.
Researchers: Y Li (Brunel).
Support: EPSRC
Duration: 2005-2008

Novel Algorithms for Gene Expression Time Series

The proposed research will examine an alternative computational framework called Simultaneous Modelling and Clustering (SMC) that will support the automatation of the gene expression time series analysis process. The SMC would cluster gene expression variables by scoring a candidate cluster on the predictive ability of a model that is built from variables within that cluster. A balanced optimisation strategy will be developed to allow quality models to be generated while managing to converge quickly. Novel scalable algorithms will be proposed to analyse gene expression time series involving thousands of variables. A systematic evaluation of the SMC methods will be performed on a variety of virus and host interaction gene expression time series using bioinformatics resources such as VIDA and BIOMAP. Parallel algorithms will be designed for running on a computer farm to speed up the process. A collection of software tools for modelling gene expression time series will be made available to life scientists.
Researchers: X Liu, M Hirsch, S Swift and A Tucker (Brunel), P Kellam (UCL), N Martin (Birkbeck), C Orengo (UCL).
Support: BBSRC
Duration: 2005-2008

DNA Microarray Data Analysis and Modelling: An Integrated Approach

This research programme will combine the expertise of a consortium of biologists, control engineers, computer scientists, mathematicians and statisticians to develop and validate a broad range of mathematical modelling tools for DNA microarray analysis, a key area of post-genomic research. The analysis will focus mainly on large robust microarray datasets obtained from a model bacterium, Streptomyces coelicolor, and a simple model eukaryotic system, herpes virus lytic replication. Our goal is to produce generic tools for modelling changes in cellular transcriptomes following induction of defined cellular processes and gain new biological insight into the gene regulation networks of the two representative model systems. These technologies and appropriate training will be made available to the wider UK functional genomics community.
Researchers: X Liu and V Vinciotti (Brunel), P Kellam (University College London), D Broomhead, M Muldoon (Manchester University), D Lowe and I Nabney (Aston University), E Wit, M Titterington, I Molchanov, A Nobile and K Vass (Glasgow University), C Smith (Surrey University), and O Wolkenhauer (Rostock University).
Support: The BBSRC/EPSRC Exploiting Genomics Initiative.
Duration: 2003-2008
Web: The MARIE Project

Integrating Transcriptomics and Structural Data to Reveal Protein Functions

The main aim of the project is to develop original methodologies for predicting the functions of uncharacterised genes. Our approach will be to integrate gene expression data with protein family, function and pathway or process data. Connection of these two important fields will significantly enhance the value of both datasets. The project will be developed using real biological data from existing projects using both Affymetrix and spotted array platforms. Two specific biological questions with direct implications for human health and well-being are addressed: the DNA damage response and the cellular response to viral infection. We will integrate the related data into a data warehouse and develop novel data mining protocols which use prior knowledge of protein family and functions to facilitate analysis of co-expressed genes. Finally, to cope with this new level of complex data integration, we will develop innovative visualisation tools to represent the pathways, processes and interactions suggested by data mining.
Researchers: C Orengo, S Nagl, D Jones, B Buxton and P Kellam (UCL), J Thornton and A Brazma (European Bioinformatics Institute), X Liu, S Swift and A Tucker (Brunel), M Hubank (Institute of Child Health), and N Martin (Birkbeck College).
Support: The Wellcome Trust Functional Genomics Development Initiative.
Duration: 2003-2007
Web: The BIOMAP Project

Cognitive Personalised Interfaces for Web-based Library Catalogues

Digital Libraries make information directly available to users via both Intranets and the Internet. Recent research has demonstrated the need for digital libraries to bridge gaps between the producers and the users of the information by providing personalised services. In particular, Web-based library catalogues serves as the major medium for such engagement. To this end, the project aims to develop a personalised Web-based library catalogue to accommodate users individual differences, especially cognitive styles. Empirical studies will be conducted to examine the effects of cognitive style on users information seeking, and generate guidelines to help designers understand how users with different cognitive styles interact with the personalised Web-based library catalogues.
Researchers: S Chen, E Frias-Martinez, X Liu, R Macredie (Brunel), and G Magoulas (Birkbeck).
Support: AHRC
Duration: 2003-2007

Multiobjective Control and Filtering for Nonlinear Stochastic Systems with Variance Constraints

The principal motivation for this proposed project is threefold. First, many engineering systems have performance requirements naturally stated in terms of the upper bounds on the steady-state variance values. Secondly, a satisfactory engineering system should possess multiple desired performances such as exponential stability, good steady-state behaviour, robustness to modelling uncertainties, acceptable disturbance rejection attenuation level, reliability, etc. Thirdly, owing to the advances in digital computers and the complexity of modelling, filtering and control for nonlinear stochastic systems have been developed and used in numerous applications. In this research project, the variance-constrained multiobjective stochastic control and filtering theory will be studied extensively for a class of nonlinear stochastic systems. Appropriate time-domain state-space design approaches and criteria will be investigated in detail based on stochastic analysis and nonlinear system theory. Numerical analysis, digital simulations and experiments for a class of engineering (biological) systems will be conducted to demonstrate the usefulness and applicability of the obtained design methods.
Researchers: Z Wang and F Yang (Brunel).
Support: EPSRC
Duration: 2003-2006

Explaining Multivariate Time Series to Detect Early Problem Signs

There has been a limited amount of work on the learning of explanation models directly from multivariate time series (MTS) data. This type of research is especially important for those applications where there is a wealth of MTS data but there is no well-established domain theory or rich body of relevant domain experience, and where detecting potential problems at an early stage is crucial. Over the last few years we have researched the issues related to the learning of such models and made important progress which calls for further investigation. This project aims to extend our current work into a coherent computational framework that is able to produce reliable and timely explanations from MTS data. This will be achieved by developing a number of advanced methods for learning efficient and reliable MTS explanation models, by integrating these methods into an effective computational framework, and by testing this framework on a variety of synthetic and real-world MTS, including DNA microarray data.
Researchers: X Liu and A Tucker (Brunel).
Support: EPSRC
Duration: 2002-2004

Analysing Virus Gene Expression Data to Understand Regulatory Interactions

This interdisciplinary project brings together four groups of researchers for addressing challenging issues in understanding virus gene interactions: the Viral Genomics and Bioinformatics Group at the Windeyer Institute of Medical Sciences, the Biomolecular Structure and Modelling Unit at UCL, the Database Technology Group at Birkbeck College, and the Intelligent Data Analysis (IDA) Group at Brunel University. It seeks to understand how to determine the genetic network of molecular interactions using DNA microarray data. A variety of clustering algorithms and related data pre-processing techniques will be used, followed by the application of novel short multivariate time series modelling techniques developed by the IDA Group. It is expected that the outcome of this research will help virologists understand better the relationship between the times when virus genes and host genes are expressed during virus replication, thereby providing important clues and insights into the virus disease process.
Researchers: X Liu and S Swift (Brunel), N Martin (Birkbeck), P Kellam (UCL Virology), and C Orengo (UCL Biochemistry).
Support: The BBSRC/EPSRC Bioinformatics Initiative.
Duration: 2001-2005

Human Factors in the Design of Adaptive Hypermedia Systems: A Cognitive Style Approach

The proposed project aims to investigate the influence of human factors on the use of Adaptive Hypermedia Learning Systems (AHLS). Current AHLS mainly focus on users prior knowledge and ignore other human factors, especially cognitive styles. We will examine how various hypermedia adaptation techniques are experienced by users with different cognitive styles. The project will integrate four aspects of research activities: (1) theoretical evaluation: to investigate and integrate existing hypermedia adaptation techniques, (2) system design: to develop an adaptive hypermedia system that can match with particular preferences of different cognitive styles, (3) empirical study: to examine the effects of different cognitive styles on learning strategies in the adaptive hypermedia system, and (4) development of guidelines: to provide guidelines about the circumstances under which particular adaptive techniques are most effective. The project can enhance the understanding of human factors in the use of adaptive hypermedia systems and help designers to develop adaptive hypermedia systems that can accommodate individual differences.
Researchers: S Chen, T Mitchell, X Liu and R Macredie (Brunel).
Support: EPSRC
Duration: 2001-2004

Bioinformatics to Discover New Animal and Human Pathogens

Bioinformatics techniques are powerful tools in assigning function to newly-identified proteins, but have not yet been widely used in virological research. We will create a virus database containing data on virus sequences, structures and derived secondary data such as conserved sequence motifs. This will provide a platform for complex analysis of viral protein families. Cross genome analysis and data mining techniques will be developed to identify novel relationships between the virus data and to assign functional information to viral families. In addition, existing search strategies based on conserved virus protein motifs will be adapted to interrogate Expression Sequence Tagged (EST) databases to identify new viral proteins associated with human diseases.
Researchers: P Kellam and M Mar Alba (UCL Virology), C Orengo (UCL Biochemistry), X Liu (Brunel) and N Martin (Birkbeck).
Support: The BBSRC/EPSRC Bioinformatics Initiative.
Duration: 1999-2003

Modelling Short Multivariate Time Series

Many statistical Multivariate Time Series (MTS) modelling methods place constraints on the minimum number of observations in the dataset, and require distribution assumptions to be made regarding the observed time series, e.g. the maximum likelihood method for parameter estimation. To date, we have developed a fast and approximate method based on evolutionary programming techniques to locate variables that are highly correlated within high-dimensional MTS. We have also demonstrated the promises of automated model order selection and parameter estimation using genetic algorithms. Specifically the method bypassed the size restrictions of the statistical methods, made no distribution assumptions, and also located the order and associated parameters as a whole step. The proposed research will extend the current work on modelling MTS data into a coherent methodology for forecasting purposes. This will be achieved by developing methods for model selection based on the current variable selection work, by improving the existing methods for model order selection and parameter estimation, and by integrating the above into an effective forecasting methodology.
Researchers: X Liu and S Swift (Brunel).
Support: EPSRC
Duration: 2000-2001

Causal Modelling for Time Series Data

One of the central objectives for this project is to generate "causal explanations of events" for mutlivariate time series data, recognised as especially valuable for process industries. The time series data used came from BP's Petroleum Refinery at Grangemouth in Scotland. In reviewing the refinery operating data, process engineers often come across trends with unexpected characteristics. In many cases, these anomalous events have a significant adverse economic impact, whether in terms of reduced yield, excessive equipment stress, or violation of environmental constraints. The identification of such events is important, but of greater importance still are adequate causal explanations of them, which could then be used to modify operating practices, retrain operators (whose inappropriate actions might have caused the events), conduct anticipatory planning, etc. Various approaches to the learning of causal models from multivariate time-series are being investigated and early results obtained are very encouraging.
Researchers: X Liu and A Tucker (Brunel), A Ogden-Swift & A Trenchard (Honeywell Hi Spec Solutions, UK), S Harp, K Lakshminarayan & T Samad ( Honeywell Technology Center, USA) and D Campbell-Brown & B Tookey (BP-Amoco).
Support: EPSRC CASE Award; Honeywell Hi Spec Solutions; BP-Amoco, Honeywell
Duration: 1997-2000.

Haemoglobin Identification by Artificial Intelligence

The task of identifying haemoglobins is a complex one, requiring considerable domain knowledge (general biochemistry knowledge about proteins and haemoglobins, various haemoglobinopathy tests, etc) and effective analysis of different types of data. These data include those from isoelectric focusing and electrophoresis, HPLC (high performance liquid chromatography), mass spectrography, the DNA structural analysis, tests about oxygen affinity and stability of haemoglogin, and those about the patient. Work that has been achieved so far includes data cleaning, data imputation, the implementation of a haemoglobin database, a system which emulates the identification procedures followed by the experts, and early encouraging identification results. We have plans for obtaining further competitive diagnostic results, for making the identification process faster and more economical, and for developing software to assist laboratory scientists.
Researchers: X Liu and S Jami, with G Loizou (Birkbeck) and S C Davies & J S Heathorn (Central Middlesex Hospital, London), F Galacteros & H Wajcman (Henri Mondor Hospital, Paris).
Support: EPSRC CASE Award; North Thames Regional Health Authority.
Duration: 1995-2000.

IDA in Screening for Eye Diseases

Visual field testing provides the eye care practitioner with essential information regarding the early detection of major blindness-causing diseases such as glaucoma. Testing of the visual field is normally performed using an expensive, specially designed instrument whose use is mostly restricted to eye hospitals. However, it is of great importance that visual field testing be offered to subjects at the earliest possible stage. By the time a patient has displayed overt symptoms and has been referred to a hospital for eye examination, it is possible that the visual field loss is already at an advanced stage and cannot be easily treated. To remedy the situation, we have exploited personal computers (PCs) as an affordable test machine. In particular, a software-based test system has been developed using machine learning methods (e.g. neural networks and decision tree induction), an intelligent user interface and a pattern discovery model, and this system has been used in several primary care settings, including prevention of optic neuritis in Africa and opportunistic detection of glaucoma in a general practice.
Researchers: X Liu and G Cheng, with G Loizou and K Cho (Birkbeck), and J Wu, B Jones, R Wormald, F Fitzke and R Hitchings (Moorfields Eye Hospital / Institute of Ophthalmology, London).
Support: British Council for Prevention of Blindness; MRC, Taiwan Postgraduate Award; United Nations Development Programme.
Duration: 1993-

Improving Glaucoma Service by Intelligent Data Analysis

This project aims to develop an intelligent data analysis (IDA) system for normal tension glaucoma management. The capabilities of the system will include a centralised glaucoma management database, intelligent data analysis functions implemented using modern computational techniques such as genetic algorithms and statistical pattern recognition, and specific decision support tools implemented using those IDA functions and expertise of clinicians. We hope that the use of this system will assist clinicians in carrying out various aspects of their work and ultimately improve the glaucoma service in eye hospitals.
Researchers: X Liu and S Swift, with J Wu, R Hitchings, R Wormald, F Fitzke (Moorfields Eye Hospital / Institute of Ophthalmology, London).
Support: EPSRC CASE Award; Moorfields Eye Hospital.
Duration: 1996-2000

Reasoning about Outliers

Outliers are difficult to handle because some of them can be measurement or recording errors, while others may represent phenomena of interest, something significant from the viewpoint of the application domain. We have so far suggested two ways of distinguishing between phenomena of interest and measurement noise. The first attempts to model "real measurements", namely how measurements should be distributed in a domain of interest, and rejects values that do not fall within the real measurements. The other uses knowledge regarding our understanding of noisy data points instead, so as to help reason about outliers. Noisy data points are modelled, and those outliers are accepted if they are not accounted for by a noise model. New approaches are also being investigated.
Researchers: X Liu and G Cheng, with J Wu (Moorfields Eye Hospital).
Support: British Council for Prevention of Blindness; Moorfields Eye Hospital.
Duration: 1994-

Structure Characterisation and Automation for Mass Spectrometry

Correlating mass spectral data with proposed chemical structures is the major task of a large number of mass spectrometry groups around the world. This correlation task requires considerable human expertise and competent spectral data analysis. The SCAMS system has been developed to provide effective computer assistance for this complex and time-consuming task. The successful pre-processing of data, before they are used for correlation, is the most important and challenging task in SCAMS and consumed most of the project's resources. Data dimensionality reduction, feature selection, and data quality assurance are among the key pre-processing activities. To this end, a variety of domain knowledge is utilised and computationally intelligent techniques are applied, which include knowledge based systems, learning meta-knowledge from an industrial database, and advanced statistical methods. These efforts have led to the acquisition of most significant and representative data for neural networks to achieve the correlation task.
Researchers: R G Johnson, K Mannock, J Phalp and H Dettmar (Birkbeck), X Liu (Brunel), A Payne (Kodak Research Division, Harrow) and J Batt (VG BioTech, Fisons Plc, Manchester).
Support: EPSRC, Department of Trade and Industry, Kodak.
Duration: 1993-1995